Once we were finished with the tests we wanted to see how well our Corsair Vengeance RAM would overclock for us in our Sandy Bridge test system. This memory is rated for 1600MHz speed with 9-9-9-24 timings and a 2T command rate at 1.35v.
We first attempted to overclock to 1866MHz. We set DRAM voltage to 1.65 and began to increase latency timings. By the time we reached 11-11-11-27 2T, the system would successfully POST, but would lock up during Windows startup. We tried loosening timings all the way to 12-12-12-28 2T at 1.65v, but we could never successfully boot into Windows.
So, we then turned our attention to improving timings at stock 1.35v voltage and 1600MHz speed. We were able to reduce command rate to 1T and improve upon the stock 9 tRP by reducing it to 8, still at 1.35v and 1600MHz.
We ran our AIDA64, Crysis 2, and Folding@Home tests again with both of these new speeds to see if there is a demonstrable improvement from the tighter timings, and we did find some improvements. We will compare the overclocked Vengeance LP settings to its stock results as well as the overclocked settings of other kits we have tested.
Our overclock resulted in only lower latency, but clock speed remained the same. So, it’s not surprising that bandwidth results aren’t drastically different.
The latency test better illustrates the improvements from the lower timings.
Since we noticed before that RAM speeds had little effect on Crysis 2 at 1280×1024, we only re-ran our 1920×1080 Crysis 2 test with our new tighter timings.
Our results demonstrate that Crysis 2 is not as sensitive to memory latency as it is bandwidth. With the massive textures being processed thanks to the High Resolution Texture Pack, it is unsurprising bandwidth is important to performance.
Our Folding @ Home test did show an appreciable improvement in performance at the overclocked latency setting. Time per frame on our 6900 work unit decreased compared to the stock settings, resulting in a nearly 1,000 point per day increase compared to the stock results.